Life stages of wall-bounded decay of Taylor-Couette turbulence
Rodolfo Ostilla-Mónico,1,*Xiaojue Zhu,2Vamsi Spandan,2Roberto Verzicco,2,3and Detlef Lohse2,4
1School of Engineering and Applied Sciences, Harvard University, Cambridge, Massachusetts 02138, USA 2Physics of Fluids Group, Faculty of Science and Technology, MESA+ Research Institute, and J. M. Burgers
Centre for Fluid Dynamics, University of Twente, P.O. Box 217, 7500 AE Enschede, The Netherlands 3Dipartimento di Ingegneria Industriale, University of Rome “Tor Vergata”,
Via del Politecnico 1, Roma 00133, Italy
4Max Planck Institute for Dynamics and Self-Organisation, 37077 Göttingen, Germany (Received 22 April 2017; published 1 November 2017)
The decay of Taylor-Couette turbulence, i.e., the flow between two coaxial and independently rotating cylinders, is numerically studied by instantaneously stopping the forcing from an initially statistically stationary flow field at a Reynolds number of Re= 3.5 × 104. The effect of wall friction is analyzed by comparing three separate cases, in which the cylinders are either suddenly made no-slip or stress-free. Different life stages are observed during the decay. In the first stage, the decay is dominated by large-scale rolls. Counterintuitively, when these rolls fade away, if the flow inertia is small a redistribution of energy occurs and the energy of the azimuthal velocity behaves nonmonotonically, first decreasing by almost two orders of magnitude and then increasing during the redistribution. The second stage is dominated by non-normal transient growth of perturbations in the axial (spanwise) direction. Once this mechanism is exhausted, the flow enters the final life stage, viscous decay, which is dominated by wall friction. We show that this stage can be modeled by a one-dimensional heat equation, and that self-similar velocity profiles collapse onto the theoretical solution.
DOI:10.1103/PhysRevFluids.2.114601
I. INTRODUCTION
Turbulence is a classic example of a nonequilibrium phenomenon: it requires a constant energy injection as energy is constantly dissipated by viscous effects. For statistically stationary turbulence, the classical picture by Richardson [1] and Kolmogorov [2–4] is that energy is injected at the larger scales, cascades down to smaller scales, and is dissipated by viscosity. In this picture, only the large scales are dependent on the boundary conditions, the forcing, and the geometry of the flow. The intermediate and small scales are assumed to be homogeneous and isotropic, and thus the energy dissipation mechanisms are postulated to be universal and self-similar.
Once the forcing is turned off, turbulence decays as energy is no longer injected into the system, but it is still dissipated. Decaying turbulence has been employed to study not only unforced systems but also the mechanisms for energy dissipation in statistically stationary turbulence, including the dissipation anomaly, a cornerstone of every theory of turbulence [5]. Historically, studies have focused on the decay of statistically homogeneous isotropic turbulence (HIT), performed both experimentally (using grid-induced turbulence) [6,7] or numerically (using a triply periodic box forced randomly) [8]. Some theoretical power laws for the decay of vorticity and kinetic energy can be obtained both from dimensional analysis and from a Navier-Stokes equation-based mean field theory [9], and are in good agreement with available data [6,7,10]. However, the dependence of decay on initial conditions [11–14] and the degree of self-similarity of the decay have been debated in the literature [8,15–17].
Real systems, however, are very far from homogeneous and isotropic. Most flows in technology and industry are wall bounded and thus anisotropic. A considerable part of the energy dissipation occurs next to the walls. And even in geo- and astrophysical flows, where walls might be absent, strong anisotropies still exist, as in accretion disks or mantle convection.
Studies of turbulence decay outside HIT are more limited [18]. Taylor-Couette (TC) flow, i.e., the turbulent flow between two concentric and independently rotating cylinders [19], is an ideal system to study the decay of wall-bounded turbulence because it is a closed system and the cylindrical geometry allows for experiments with relatively small end-effects and more modest sizes, unlike channel or pipe flow. While TC had already been employed to study decay of puffs at low Reynolds numbers [20], or turbulence decay due to linear stabilization [21], the experiment by Verschoof et al. [22] focused on the fully unstable turbulent regime, at Reynolds numbers of Re∼ O(106). The cylinders were stopped, and the turbulence was allowed to decay. The decay did not follow a pure power law like HIT, but instead decayed faster due to the viscous drag applied by the walls. Nonetheless, the decay was found to be self-similar. However, the considerable inertia of the massive metal cylinders causes the braking to take a substantial amount of time (approximately 3000 eddy turnover times), during which the decay could not be measured. The experiment also has secondary flows due to residual cooling effects, which become important at later times and cause deviations from the models in Ref. [22].
In this paper, we numerically simulate Taylor-Couette turbulence, analogous to the experiments of Ref. [22], but now we instantaneously stop the cylinders (which is of course possible in numerical simulations) to better understand the initial stages of the decay, and run the simulation until the fully viscous final stage dominates the system dynamics. In addition, we can remove the effect of wall friction by making the walls stress-free, and thus we can clearly separate the role of the walls from other decay mechanisms in the flow. We find that the decay regime observed and discussed in Ref. [22] corresponds to an intermediate stage of decay. We do not find the proposed self-similar regime until the last life stage which can be described as a quenching problem. Prior to this last viscosity-dominated stage, two earlier life stages of decay are observed, which are dominated by linear instabilities and non-normal transient growth, respectively.
II. NUMERICAL DETAILS
We simulate the incompressible Navier-Stokes equations using a second-order energy-conserving centered finite difference scheme, which treats the viscous terms semi-implicitly [23,24]. The initial starting field corresponds to a TC system driven by pure inner cylinder rotation in an inertial reference frame. To reduce dispersive errors in the simulations, we simulate the system in a rotating frame such that the velocities at both cylinders are always equal and of opposite signs (±U/2). Due to this reference frame change, a Coriolis force naturally arises [25]. The initial shear Reynolds number was taken as Res= Ud/ν = 3.52 × 104, where ri(ro) is the inner (outer) cylinder radius, d the gap width
d = ro− ri, and ν the kinematic viscosity of the fluid. Periodic boundary conditions are employed
in the axial direction. The system has a radius ratio of η= ri/ro= 0.909 and an aspect ratio of
= Lz/(ro− ri)= 4, where Lzis the axial periodicity length. This geometry was chosen such that
the system had two fixed roll pairs (i.e., four rolls) and curvature effects are limited. A rotational (azimuthal) symmetry of order nsym= 10 was forced on the system to reduce computational costs. This means that the system is 4π half-gaps long in the streamwise direction at its smallest extent. A resolution of Nθ× Nr× Nz= 1024 × 512 × 1024 was used; the grid nodes were uniformly
distributed in the axial and azimuthal directions, while a clipped Chebychev distribution was used for the radial direction to cluster points near the walls. This is equivalent to the resolutions used for previous studies [26].
The simulations were run until a statistically stationary state was reached, where two roll pairs could be seen in the velocity fields, and the torque at both cylinders was equal when temporally averaging over a sufficiently long time. We simulate three cases: for the first case, which we denote as sudden stop as in the experiment (SS-EXP), we instantaneously stop the inner cylinder at t = 0,
and allow the turbulence to decay. The main difference between the simulation and the experiment is that we stop the cylinders instantaneously, while the stopping time of the experiment by Verschoof et al. [22] is around 12 s. Even if this is still orders of magnitude smaller than the viscous time (d2/ν= 6 × 103s), it corresponds to over 3000 large eddy turnover times, as mentioned previously (d/U = 3.2 × 10−3s).
To eliminate as much as possible the effects of the wall, we run a second case, denoted by sudden disengagement (SD), in which we suddenly change the boundary condition at the wall to stress-free. The flow is no longer forced by shear, and thus the turbulence decays. Finally, we run an intermediate, third case, denoted by (SS-V0) where we set the velocity of both cylinders to the mean velocity, i.e., zero in the rotating frame. While in the experiment, and in the SS-EXP simulation, the inner cylinder is brought to the same velocity as the outer cylinder, and there is a substantial mean momentum which must be transported from bulk to walls, this is not the case in the SS-V0 case because the mean momentum with respect to the cylinders is close to zero. In both the SS-V0 case and the SD case, the main decay is that of velocity fluctuations while in the SS-EXP case it is of the mean azimuthal (streamwise) velocity. Thus, we expect the SS-V0 case to behave in an intermediate way between the SD case and the SS-V0 case.
As turbulence decays, the computational method is switched for efficiency. At ˜t= tU/d ≈ 125 for the SS-V0 case, and ˜t≈ 325 for the SD and SS-EXP cases, the flow fields were down-sampled to a resolution of 5123, and the treatment of the viscous terms in the homogeneous directions was made implicit. This allowed the time step to increase to ˜t= 1, dramatically reducing the computational cost. The simulations were then advanced in time up to ˜t= 7500 (SS-V0/SD) and ˜t= 15000 (SS-EXP), late enough to be in the asymptotic viscous stage.
III. RESULTS
Figure1shows the temporal evolution of the average kinetic energy of the azimuthal velocity (12u2
θ) and the wind kinetic energy (12ur2+ u2z) for all cases, while Fig.2shows the temporal
evolution of the total kinetic energy dissipation rate =12ν∂iuj2, where · · · denotes the
averaging operator and the entire fluid volume. On the right panel of Fig.2, experimental data from [22] are also added for comparison. This data only matches the top temporal axis (viscid units).
No overarching behavior or power law which describes either the dissipation or the evolution of the kinetic energy can be seen. Instead, several different life stages of the decay are revealed, which we detail here. In Fig.1, the main difference between the SS-EXP and the other two cases can be clearly seen: the azimuthal kinetic energy (in the bulk) is several orders of magnitude larger than in the other cases. Throughout the entire decay, it is significantly larger than the wind kinetic energy, in line with what was observed by Ref. [22]. As we will see, this dominates the physics of the decay. The black curve in the right panel of Fig.2shows a much faster decay in the dissipation rate (when measured in viscous time units) than any of the numerical cases simulated; the reasons for this will be explored later.
The stages of decay are illustrated by Figs.3and4, which shows a visualizations of the azimuthal velocity at different time instants for the SS-V0 and SS-EXP cases, from the start of the simulation. Aside from the magnitude of the velocity being different, we can also see wide variation in the flow topology. A similar flow topology as the one seen for SS-V0 is seen for the SD case, too.
We briefly describe the stages for the SS-V0 and SD cases: during the first life stage, which takes place between ˜t= 0 to ˜t ≈ 10 [corresponding to Fig. 3(b)], the large-scale rolls remain in motion, as they are still being fed from the perturbations inside the boundary layers. After this, the flow undergoes nonlinear non-normal transient growth, as in an ordinary shear flow. Finally, in the last, asymptotic stage, (shown in the rightmost panels), viscous diffusion dominates and the flow becomes homogeneous in the azimuthal and axial directions.
The SS-EXP case shows different behavior, as the large scale roll remains being fed throughout the entire simulation by the outer cylinder, while the inner cylinder undergoes a rapid change from feeding the roll through a centrifugal instability for ˜t < 0 to being centrifugally stable for ˜t > 0.
10−2 100 102 104 tU/d 10−8 10−6 10−4 10−2 100 u 2 θΩ 10−5 tν/d102−3 10−1 0 2500 5000 7500 10000 tU/d 10−8 10−6 10−4 10−2 100 u 2 θΩ 0.0 0.1 tν/d0.2 0.3 2 10−2 100 102 104 tU/d 10−12 10−10 10−8 10−6 10−4 10−2 u 2 r+ u 2 zΩ 10−5 tν/d102−3 10−1 0 2500 5000 7500 10000 tU/d 10−12 10−10 10−8 10−6 10−4 10−2 u 2 r+ u 2 zΩ 0.0 0.1 tν/d0.2 0.3 2
FIG. 1. Top left: Temporal evolution of the kinetic energy in the azimuthal velocity. Bottom left: Temporal evolution of the wind kinetic energy. The figures show the same data in semilogarithmic scale (right) and double logarithmic scale (left). Symbols: SS-EXP (green), SD (red), and SS-V0 (blue).
This is especially noticeable in the second panel of Fig.4, where remnants of ejection zones of the outer cylinder are clearly visible, but the flow topology at the inner cylinder has changed. After this phase, the flow loses the clear axial signature of the rolls (third panel), but the strong boundary layer asymmetry remains in place, as is evident in the fourth panel. At even later times, shown in the last
10−2 100 102 104 tU/d 104 106 108 1010 1012 4/ν 3 10−5 tν/d102 −3 10−1 0 2500 5000 7500 10000 tU/d 104 106 108 1010 1012 4/ν 3 0.0 0.1 tν/d0.2 0.3 2
FIG. 2. Temporal evolution of the energy dissipation. The figures show the same data in semilogarithmic scale (right) and double logarithmic scale (left). Symbols: SS-EXP (green), SD (red), SS-V0 (blue), experimental data from Ref. [22] (right, data matches top temporal axis only).
(a) (b) (c) (d)
FIG. 3. Pseudocolor visualization of the azimuthal velocity in a constant-azimuth cut for the SS-V0 case at four different times (left to right): ˜t= 0, 10, 75, and 5000.
panel, the flow also enters the asymptotic stage, where it becomes practically homogeneous in the azimuthal and axial directions, analogous to the right panel of Fig.3.
No significant axial signature was observed in Ref. [22] after the long stopping time of the cylinders, which provides some indication that their measurements start at a time where the flow approximately behaves as in the third panel of Fig.4.
Due to the less clean delimitation between stages in the SS-EXP case, we first focus on the three stages of the SS-V0 and SD cases. During the first life stage, which takes place between ˜t= 0 to ˜t≈ 10 [corresponding to Fig.3(b)], the large-scale rolls remain in motion, as they are still fed from the perturbations inside the boundary layers. Figure1shows that the wind kinetic energy remains constant (horizontal) between ˜t= 0 and ˜t ≈ 10, while the kinetic energy of the azimuthal velocity
(a) (b) (c) (d) (e)
FIG. 4. Pseudocolor visualization of the azimuthal velocity in a constant-azimuth cut for the SS-EXP case at five different times (left to right): ˜t= 0, 10, 100, 1000, and 17000.
0.0 0.2 0.4 0.6 0.8 1.0 (r − ri)/d 10−4 10−3 10−2 10−1 100 u 2 θz, θ 0.0 0.2 0.4 0.6 0.8 1.0 (r − ri)/d 10−2 10−1 100 u 2 θz, θ
FIG. 5. Left: Azimuthally and axially averaged azimuthal kinetic energy dissipation for the SS-V0 case, and ˜t= 0 (blue), ˜t = 10 (orange), ˜t = 30 (green), and ˜t = 50 (red). Right: The same for SS-EXP case and ˜t= 0 (blue), ˜t = 10 (orange), and ˜t = 50 (green).
decreases by more than an order of magnitude. The different behavior of velocity components highlights the anisotropic character of the decay.
We note that the time scale of this first stage is the same for both the SS-V0 and the SD cases, highlighting that the rolls drain energy much more efficiently than the wall. A simple scaling estimate for the diffusive time across the boundary layer of the available kinetic energy to the wall is tν,BL= λ2ω/ν. If we estimate the boundary layer thickness λωas d/(2N uω), with N uω the
generalized Nusselt number from the initial turbulent simulation and equal to N uω≈ 25 [26], we
obtain an estimate for tν,BLU/d≈ 200, an order of magnitude larger than the time scale seen. This
stage is not seen in Fig.1for the SS-EXP case, because the kinetic energy is dominated by the bulk and not by the boundary layers. From this we would also not expect this stage to be present in the experiment.
Further proof of this is seen in Fig.5, which shows the distribution of the azimuthal kinetic energy at the start and end of the first stage, and at similar times for the SS-EXP case. At the start of the decay, energy is concentrated near the boundary layers. Once this energy is drained by the rolls, they fade away. The boundary layers grow, and the fluctuation maximum moves away from the walls. Energy is redistributed between the existing structures, and the flow topology changes from Fig.3(b)
to3(c). Counterintuitively, the decay of azimuthal kinetic energy is nonmonotonic, and in Fig.5, we see an increase in azimuthal kinetic energy at around ˜t≈ 15 which is drawn from the wind kinetic energy during the redistribution, up to a maximum at ˜t≈ 40. Wall friction plays a secondary role during this stage. The main player draining the available energy in the boundary layer is the rolls. The wall friction comes in as a small correction, making the SS case undergo a transition from the first life stage to the second life stage slightly faster than in the SD case. Nonmonotonic behavior can be seen for both SS and SD cases. For the SS-EXP case, no redistribution is seen. The energy levels for the bulk are much higher and this dominatesu2θ.
Then the second life stage is entered, which takes place between ˜t≈ 10 and ˜t ≈ 500. With the rolls fade away, perturbations are amplified non-normally. A detailed discussion of this non-normal nonlinear mechanism is found in Ref. [27]. Here it is sufficient to state that perturbations in the axial (spanwise) direction tend to grow faster than those in other directions. As is typical for this type of instability, the growth of these perturbations is transient, and after several time units they decay. For the SS-V0 case, the wind kinetic energy drops to a minimum aroundu2
≈ 10−5and it remains
at that level until the end of this regime, while for the SD case the wind kinetic energy does not reach a plateau but instead oscillations appear throughout the entire decay. The SS case appears to drain the available energy of the many non-normal modes in a shorter amount of time, while this draining occurs over a longer time scale for the SD case. The signature of non-normal transient
(a) (b)
FIG. 6. Pseudocolor visualization of the azimuthal velocity in a constant-radius cut at the mid-gap (r= ra) for the SS-V0 case at two different times ˜t= 75 (top) and 1000 (bottom).
decay can be clearly seen for the wind kinetic energy in the SD and SS-EXP cases up to the end of the simulation. When one of the non-normal modes grows, amplifies transiently, and decays, it imprints the oscillations seen in the wind kinetic energy for ˜t > 100.
Due to the non-normal transient growth and dissipation of axially oriented perturbations, the flow becomes practically homogeneous in the axial direction for ˜t > 500 in the SS-V0 case. The large-scale flow structure is now columnar. This is shown in Fig.6. The SD and SS-EXP cases do not show this, and instead show both azimuthal and axial dependency even at later times. Unlike the SS-V0 case, the marks of nonlinear transient growth seem to be present in the wind kinetic energy even up to ˜t≈ 2500 for these two cases.
The experimental data shows an even steeper decay of than the DNS in Fig.2. The instantaneous Reynolds number is still significantly higher in the experiment than in the simulations even for later times. Significant nonlinear non-normal transient growth can still happen, which means an overall faster draining of energy and an overall faster decay rate (when looking at the flow in viscid time units). This is reflected in a steeper decay of .
The role of wall friction is minor in the transition between the second and third stage, as this happens when transient non-normal growth is exhausted. Once either the nonlinear transient growth mechanism is drained or the wind kinetic energy is sufficiently small that its effect cannot be felt, the decay enters its last life stage, in which the decay is dominated by viscosity.
In this last life stage the energy dissipation rate behaves in a quasiexponential manner. The wind kinetic energy is negligible and we can expect the azimuthal velocity to behave like a passive scalar with diffusion, i.e., a heat equation. With the boundary and initial conditions, the problem is analogous to the quenching of a cylindrical annulus, whose solution is given by uθ(r,t)=
nAnJ0(λnr) exp(−νλ2nt), where J0(r) are Bessel functions of the zeroth order and λntheir associated eigenvalues, and Andepend on the initial conditions [28]. If t is large, only the
slowest-decaying function with the smallest eigenvalue has significant energy. The energy dissipation behaves exponentially in time, and self-similarity comes out naturally.
To demonstrate that for very large times (beyond ˜t > 5000 for SS-V0 and ˜t > 15000 for SS-EXP) the asymptotic self-similar decay regime is entered, in Fig.7we show several azimuthal velocity profiles, normalized by their mean value. For the SS-V0 case, the normalized velocity profiles at times ˜t= 5000 to the end of our simulation ˜t = 7500 (corresponding to tν/d2 ≈ 0.2) collapse, and agree with the theoretically calculated fundamental Bessel function, while the earlier normalized profiles between ˜t≈ 1000 and ˜t ≈ 5000 show slight deviations, as the asymptotic stage has not yet been reached. The normalized profile around ˜t≈ 500 is very different, as it is in the non-normal transient growth state. For the SS-EXP case, the asymptotic, self-similar stage is only reached for ˜t≈ 15 000, corresponding to tν/d2≈ 0.25.
When comparing these profiles to the proposed self-similarity in Ref. [22], we find that they are similar at intermediate times: a flat profile in the center, indicating strong turbulent mixing, and a
0.0 0.2 0.4 0.6 0.8 1.0 (r − ri)/d −0.5 0.0 0.5 1.0 1.5 2.0 uθ z, θ / uθ Ω 0.0 0.2 0.4 0.6 0.8 1.0 (r − ri)/d 0.0 0.5 1.0 1.5 uθ z, θ / uθ Ω
FIG. 7. Left: Axially and azimuthally averaged normalized azimuthal velocity profiles for the SS-V0 case at ˜t= 500 (blue), ˜t = 1000 (green), and ˜t = 5000 (orange), as well as the theoretical solution based on Bessel functions (black dashed). Right: The same for the SS-EXP case at times ˜t= 10 (blue), ˜t = 100 (orange), ˜t= 1000 (green), ˜t = 5000 (red), and ˜t = 15000 (purple).
steeper boundary layer at the outer cylinder, due to the centrifugal instability at the inner cylinder and the centrifugal stability at the outer cylinder. This intuition coincides with the fact that the largest reported profile at t = 100s is only tν/d2≈ 0.03 in viscous units. The proposed self-similarity is shown for a small interval of time (0.001 < tν/d2<0.03), and may be just be a product of the relatively small time interval.
IV. CONCLUSIONS AND OUTLOOK
In summary, in this paper, we have studied the decay of Taylor-Couette turbulence, focusing also on the early and late stages of decay that could not be observed in Ref. [22]. We have shown that three distinct life stages are seen where different decay mechanisms dominate. In the first life stage, the energy decays mainly through the linearly unstable modes, i.e., the rolls. After this, the available energy is directed towards non-normal modes which sustain transient growth. During the transition between stages, a redistribution of energy between structures occurs, anduθ2 increases. While
this can only be seen if the flow inertia is small, it highlights that the decay process is anisotropic and be temporally non-monotonic. Finally, the decay becomes purely viscous, in spite of the relatively high instantaneous Reynolds number, and self-similarity can be observed. We have shown that no simple power law can cover all three stages. The most unstable modes dominate the early decay; the modes decay progressively, from linearly unstable modes with roll-like instabilities, to non-normal transient growth, to self-similar decay, which behaves like a quenching problem.
This progression of decaying modes from more to less unstable can shed light onto how real-world unforced turbulence decays. Our findings—that anisotropic flows can decay in a viscid way even at moderate Reynolds numbers, that redistribution of energy between structures can lead to nonmonotonic behavior, and that wall-friction is a secondary mechanism in the initial decay—can provide insight into the decay of geo- and astrophysically relevant systems. We expect these findings to be widely applicable even if geo- and astro-physical flows are unconfined, unlike Taylor-Couette flow. Further insights into decay and the crossover between the first two stages can be provided by studying the decay in linearly stable Taylor-Couette flow [29], or in plane Couette flow, i.e., the flow between two parallel plates, which could make the first stage of decay less important. The TC geometries studied have very limited curvature, and its (de)stabilizing role in reducing the non-normal transient growth can be explored. Finally, another research line which can be explored are studies of the decay of thermal turbulence [30], and the relation between decaying TC flow and decaying Rayleigh-Bénard (RB) convection [31,32] (the flow in a layer heated from below and
cooled from above). RB and TC have been called the “twins of turbulence research” [33], and RB is also commonly used as a model for geo- and astro-physical flows.
ACKNOWLEDGMENTS
We thank B. Eckhardt for fruitful and stimulating discussions and R. A. Verschoof for providing the data from Ref. [22]. This research was supported in part by the National Science Foundation under Grant No. PHY11-25915. We also gratefully acknowledge computational time for the simulations provided by SURFsara on resource Cartesius through a NWO Grant.
[1] L. F. Richardson, The supply of energy from and to atmospheric eddies,Proc. R. Soc. London Ser. A 97,
354(1920).
[2] A. N. Kolmogorov, The local structure of turbulence in incompressible viscous fluid for very large Reynolds numbers, Dokl. Akad. Nauk SSSR 30, 9 (1941), reprinted inProc. R. Soc. London A 434,9
(1991).
[3] A. N. Kolmogorov, On degeneration (decay) of isotropic turbulence in incompressible viscous liquid, Dokl. Akad. Nauk SSSR 31, 538 (1941).
[4] A. N. Kolmogorov, Dissipation of energy in locally isotropic turbulence, Dokl. Akad. Nauk SSSR 32, 16 (1941), reprinted in Proc. R. Soc. London A 434, 15 (1991).
[5] J. C. Vassilicos, Dissipation in turbulent flows,Annu. Rev. Fluid Mech. 47,95(2015).
[6] M. R. Smith, R. J. Donnelly, N. Goldenfeld, and W. F. Vinen, Decay of Vorticity in Homogeneous Turbulence,Phys. Rev. Lett. 71,2583(1993).
[7] S. R. Stalp, L. Skrbek, and R. J. Donnelly, Decay of Grid Turbulence in a Finite Channel,Phys. Rev. Lett.
82,4831(1999).
[8] W. K. George and H. Wang, The exponential decay of homogeneous turbulence,Phys. Fluids 21,025108
(2009).
[9] D. Lohse, Crossover from High to Low Reynolds Number Turbulence,Phys. Rev. Lett. 73,3223(1994). [10] L. D. Landau and E. M. Lifshitz, Fluid Mechanics (Pergamon, Oxford, 1987).
[11] P. Lavoie, L. Djenidi, and R.A. Antonia, Effects of initial conditions in decaying turbulence generated by passive grids,J. Fluid Mech. 585,395(2007).
[12] P. C. Valente and J. C. Vassilicos, The decay of turbulence generated by a class of multiscale grids,J. Fluid Mech. 687,300(2011).
[13] D. Hurst and J. C. Vassilicos, Scalings and decay of fractal-generated turbulence,Phys. Fluids 19,035103
(2011).
[14] T. Teitelbaum and P. D. Mininni, Effect of Helicity and Rotation on the Free Decay of Turbulent Flows,
Phys. Rev. Lett. 103,014501(2009). [15] P. G. Saffman,Phys. Fluids 10,1349(1967).
[16] L. Biferale, E. Calzavarini, F. Toschi, and R. Tripiccione, Universality of anisotropic fluctuations from numerical simulations of turbulent flows,Europhys. Lett. 64,461(2003).
[17] G. L. Eyink and D. J. Thompson, Free decay of grid turbulence in a finite channel,Phys. Fluids 12,477
(2000).
[18] H. Touil, J.-P. Bertoglio, and L. Shao, The decay of turbulence in a bounded domain,J. Turbulence 3,N49
(2002).
[19] S. Grossmann, D. Lohse, and C. Sun, High–Reynolds number Taylor-Couette turbulence,Annu. Rev. Fluid Mech. 48,53(2016).
[20] Daniel Borrero-Echeverry, Michael F. Schatz, and Randall Tagg, Transient turbulence in Taylor-Couette flow,Phys. Rev. E 81,025301(R)(2010).
[21] R. Ostilla-Monico, R. Verzicco, S. Grossmann, and D. Lohse, Turbulence decay towards the linearly-stable regime of Taylor-Couette flow,J. Fluid Mech. 747,1(2014).
[22] R. A. Verschoof, S. G. Huisman, R. C. A. van der Veen, C. Sun, and D. Lohse, Self-similar decay of high reynolds number Taylor–Couette turbulence,Phys. Rev. Fluids 1,062402(R)(2016).
[23] R. Verzicco and P. Orlandi, A finite-difference scheme for three-dimensional incompressible flow in cylindrical coordinates,J. Comput. Phys. 123,402(1996).
[24] E. P. van der Poel, R. Ostilla-Monico, J. Donners, and R. Verzicco, A pencil distributed finite difference code for strongly turbulent wall-bounded flows,Comput. Fluids 116,10(2015).
[25] R. Ostilla-Monico, E. P. van der Poel, R. Verzicco, S. Grossmann, and D. Lohse, Exploring the phase diagram of fully turbulent Taylor-Couette flow,J. Fluid Mech. 761,1(2014).
[26] R. Ostilla-Mónico, D. Lohse, and R. Verzicco, Effect of roll number on the statistics of turbulent Taylor-Couette flow,Phys. Rev. Fluids 1,054402(2016).
[27] S. Grossmann, The onset of shear flow turbulence,Rev. Mod. Phys. 72,603(2000).
[28] A. Prosperetti, Advanced Mathematics for Applications (Cambridge University Press, Cambridge, 2011). [29] R. Ostilla-Mónico, R. Verzicco, and D. Lohse, Turbulent taylor-couette flow with stationary inner cylinder,
J. Fluid Mech. 799,R1(2016).
[30] G. Ahlers, S. Grossmann, and D. Lohse, Heat transfer and large scale dynamics in turbulent Rayleigh-Bénard convection,Rev. Mod. Phys. 81,503(2009).
[31] B. Dubrulle and F. Hersant, Momentum transport and torque scaling in Taylor-Couette flow from an analogy with turbulent convection,Eur. Phys. J. B 26,379(2002).
[32] B. Eckhardt, S. Grossmann, and D. Lohse, Torque scaling in turbulent Taylor-Couette flow between independently rotating cylinders,J. Fluid Mech. 581,221(2007).
